MONGE-AMPE`RE EQUATIONS IN BIG COHOMOLOGY CLASSES 0 1 S. BOUCKSOM, P.EYSSIDIEUX,V.GUEDJ, A.ZERIAHI 0 2 Abstract. We define non-pluripolar products of an arbitrary number of p closed positive (1,1)-currents on a compact K¨ahler manifold X. Given a e big (1,1)-cohomology class α on X (i.e. a class that can be represented by S astrictlypositivecurrent)andapositivemeasureµonX oftotalmassequal 9 to the volume of α and putting no mass on pluripolar sets, we show that µ can bewritten in auniqueway as the top degree self-intersection in thenon- ] pluripolarsenseofaclosedpositivecurrentinα. WethenextendKolodziedj’s V approachtosup-normestimatestoshowthatthesolutionhasminimalsingu- C laritiesinthesenseofDemaillyifµhasL1+ε-densitywithrespecttoLebesgue . measure. If µ is smooth and positive everywhere, we prove that T is smooth h on the ample locus of α provided α is nef. Using a fixed point theorem we t a finally explain how to construct singular K¨ahler-Einstein volume forms with m minimal singularities on varieties of general type. [ 2 v 4 Contents 7 6 Introduction 2 3 1. Non-pluripolar products of closed positive currents 4 . 2 2. Weighted energy classes 21 1 3. Non-pluripolar measures are Monge-Amp`ere 32 8 4. L∞-a priori estimate 39 0 : 5. Regularity of solutions: the case of nef classes 43 v 6. Singular Ka¨hler-Einstein metrics 47 i X References 52 r a Bibliography 52 Date: 1 December2009. 1 2 S.BOUCKSOM,P.EYSSIDIEUX,V.GUEDJ,A.ZERIAHI Introduction The primary goal of the present paper is to extend the results obtained by the last two authors on complex Monge-Amp`ere equations on compact Ka¨hler manifolds in [31] to the case of arbitrary cohomology classes. More specifically let X be a compact n-dimensional Ka¨hler manifold and let ω be a Ka¨hler form on X. An ω-plurisubharmonic (psh for short) function is an upper semi-continuous function ϕ such that ω +ddcϕ is positive in the sense of currents, and [31] focused on the Monge-Amp`ere type equation (ω+ddcϕ)n = µ (0.1) where µ is a positive measure on X of total mass µ(X) = ωn. As is well- X known, it is not always possible to make sense of the left-hand side of (0.1), R but it was observed in [31] that a construction going back to Bedford-Taylor [4] enables in this global setting to define the non-pluripolar part of the would-be positive measure (ω+ddcϕ)n for an arbitrary ω-psh function ϕ. In the present paper we first give a more systematic treatment of these issues. Weexplainhowtodefineinasimpleandcanonicalwaythenon-pluripolar product hT ∧...∧T i 1 p ofarbitraryclosed positive(1,1)-currents T ,...,T with1 ≤ p ≤ n. Theresulting 1 p positive (p,p)-current hT ∧...∧T i puts no mass on pluripolar subsets. It is also 1 p shownto beclosed (Theorem 1.8), generalizing aclassical resultof Skoda-ElMir. We relate its cohomology class to the positive intersection class hα ·····α i ∈Hp,p(X,R) 1 p of the cohomology classes α := {T } ∈ H1,1(X,R) in the sense of [13, 14]. In j j particular we show that hTni ≤ vol(α) where the right-hand side denotes the X volume of the class α := {T} [11], which implies in particular that hTni is non- R trivial only if the cohomology class α is big. Animportantaspectofthepresentapproachisthatthenon-pluripolarMonge- Amp`ere measure hTni is well defined for any closed positive (1,1)-current T. In the second section we study the continuity properties of the mapping T 7→ hTni: it is continuous along decreasing sequences (of potentials) if and only if T has full Monge-Amp`ere mass (Theorem2.17), i.e. when hTni = vol(α). ZX We prove this fact defining and studying weighted energy functionals in this general context, extending the case of a Ka¨hler class [31]. The two main new featuresareageneralized comparisonprinciple(Corollary 2.3)andanasymptotic criterion to check whether a current T has full Monge-Amp`ere mass (Proposition 2.19). In the third part of the paper we obtain our first main result (Theorem 3.1): Theorem A. Let α ∈ H1,1(X,R) be a big class on a compact Ka¨hler manifold X. If µ is a positive measure on X that puts no mass on pluripolar subsets and MONGE-AMPE`RE EQUATIONS IN BIG COHOMOLOGY CLASSES 3 satisfies the compatibility condition µ(X) = vol(α), then there exists a unique closed positive (1,1)-current T ∈ α such that hTni= µ. The existence part extends the main result of [31], which corresponds exactly to the case where α is a Ka¨hler class. In fact the proof of Theorem A consists in reducingto the Ka¨hler case via approximate Zariski decompositions. Uniqueness is obtained by adapting the proof of S. Dinew [25] (which also deals with the Ka¨hler class case). When the measure µ satisfies some additional regularity condition, we show how to adapt Ko lodziej’s pluripotential theoretic approach to the sup-norm a priori estimates [35] to get global information on the singularities of T. Theorem B. Assume that the measure µ in Theorem A furthermore has L1+ε density with respect to Lebesgue measure for some ε > 0. Then the solution T ∈ α to hTni= µ has minimal singularities. Currents with minimal singularities were introduced by Demailly. When α is a Ka¨hler class, the positive currents T ∈ α with minimal singularities are exactly those with locally bounded potentials. When α is merely big all positive currents T ∈ α will have poles in general, and the minimal singularity condition on T essentially says that T has the least possible poles among all positive currents in α. Currents with minimal singularities have in particular locally bounded potentials ontheample locus Amp(α) of α, which isroughly speakingthelargest Zariski open subset where α locally looks like a Ka¨hler class. Regarding local regularity properties, we obtain the following result. Theorem C. In the setting of Theorem A, assume that µ is a smooth strictly positive volume form. Assume also that α is nef. Then the solution T ∈ α to the equation hTni = µ is C∞ on Amp(α). Theexpectation is ofcoursethatTheoremCholdswhetheror notαis nef,but we are unfortunately unable to prove this for the moment. It is perhaps worth emphasizingthat currents with minimalsingularities can have a non empty polar set even when α is nef and big (see Example 5.4). In the last part of the paper we consider Monge-Amp`ere equations of the form h(θ+ddcϕ)ni = eϕdV (0.2) where θ is a smooth representative of a big cohomology class α, ϕ is a θ-psh function and dV is a smooth positive volume form. We show that (0.2) admits a unique solution ϕ such that eϕdV = vol(α). Theorem B then shows that ϕ X has minimal singularities, and we obtain as a special case: R Theorem D. Let X be a smooth projective variety of general type. Then X admits a unique singular K¨ahler-Einstein volume form of total mass equal to vol(K ). In other words the canonical bundle K can be endowed with a unique X X non-negatively curved metric e−φKE whose curvature current ddcφKE satisfies h(ddcφ )ni= eφKE (0.3) KE 4 S.BOUCKSOM,P.EYSSIDIEUX,V.GUEDJ,A.ZERIAHI and such that eφKE = vol(K ). (0.4) X ZX The weight φ furthermore has minimal singularities. KE Since the canonical ring R(K ) = ⊕ H0(kK ) is now known to be finitely X k≥0 X generated [7], this result can be obtained as a consequence of [28] by passing to the canonical model of X. But one of the points of the proof presented here is to avoid the use of the difficult result of [7]. The existence of φ satisfying (0.3) and (0.4) was also recently obtained by KE J. Song and G. Tian in [43], building on a previous approach of H. Tsuji [49]. It is also shown in [43] that φ is an analytic Zariski decomposition (AZD KE for short) in the sense of Tsuji, which means that every pluricanonical section σ ∈ H0(mK ) is L∞ with respect to the metric induced by φ . The main X KE new information we add is that φ actually has minimal singularities, which KE is strictly stronger than being an AZD for general big line bundles (cf. Proposi- tion 6.5). Acknowledgement. We would like to thank J.-P.Demailly and R.Berman for several useful discussions. 1. Non-pluripolar products of closed positive currents In this section X denotes an arbitrary n-dimensional complex manifold unless otherwise specified. 1.1. Plurifine topology. The plurifine topology on X is defined as the coarsest topologywithrespecttowhichallpshfunctionsuonallopensubsetsofX become continuous (cf. [4]). Note that the plurifinetopology on X is always strictly finer than the ordinary one. Since psh functions are upper semi-continuous, subsets of the form V ∩{u > 0} withV ⊂ X openandupshonV obviouslyformabasisfortheplurifinetopology, and u can furthermore be assumed to be locally bounded by maxing it with 0. When X is furthermore compact and Ka¨hler, a bounded local psh function defined on an open subset V of X can be extended to a quasi-psh function on X (possibly after shrinking V a little bit, see [30]), and it follows that the plurifine topologyonX canalternativelybedescribedasthecoarsesttopologywithrespect to which all bounded and quasi-psh functions ϕ on X become continuous. It therefore admits sets of the form V ∩{ϕ > 0} with V ⊂ X open and ϕ quasi-psh and bounded on X as a basis. 1.2. Local non-pluripolar products of currents. Let u ,...,u be psh func- 1 p tionsonX. Iftheu ’sarelocallybounded,thenthefundamentalworkofBedford- j Taylor [3] enables to define ddcu ∧...∧ddcu 1 p MONGE-AMPE`RE EQUATIONS IN BIG COHOMOLOGY CLASSES 5 on X as a closed positive (p,p)-current. The wedge product only depends on the closed positive (1,1)-currents ddcu , and not on the specific choice of the j potentials u . j Avery importantpropertyof thisconstruction isthat itis local in the plurifine topology, in the following sense. If u , v are locally bounded psh functions on X j j and u = v (pointwise) on a plurifine open subset O of U, then j j 1 ddcu ∧...∧ddcu = 1 ddcv ∧...∧ddcv . O 1 p O 1 p This is indeed an obvious generalization of Corollary 4.3 of [4]. In the case of a possibly unbounded psh function u, Bedford-Taylor have ob- served (cf. p.236 of [4]) that it is always possible to define the non-pluripolar part of the would-be positive measure (ddcu)n as a Borel measure. The main issue however is that this measure is not going to be locally finite in general (see Example 1.3 below). More generally, suppose we are trying to associate to any p-tupleu ,...,u of pshfunctionson X a positive (a priori not necessarily closed) 1 p (p,p)-current hddcu ∧...∧ddcu i 1 p putting no mass on pluripolar subsets, in such a way that the construction is local in the plurifine topology in the above sense. Then we have no choice: since u coincides with the locally bounded psh function max(u ,−k) on the plurifine j j open subset O := {u > −k}, k j j \ we must have 1 h ddcu i = 1 ddcmax(u ,−k), (1.1) Ok j Ok j j j ^ ^ Note that the right-hand side is non-decreasing in k. This equation completely determinesh ddcu isincethelatterisrequirednottoputmassonthepluripolar j j set X − O = {u = −∞}. k Vk Definition 1.1. If u ,...,u are psh functions on the complex manifold X, we S 1 p shall say that the non-pluripolar product h ddcu i is well-defined on X if for j j each compact subset K of X we have V sup ωn−p∧ ddcmax(u ,−k) < +∞ (1.2) j k ZK∩Ok j ^ for all k. Hereω isanauxiliary (strictly) positive(1,1)-form onX withrespecttowhich massesarebeingmeasured,theconditionbeingofcourseindependentofω. When (1.2) is satisfied, equation (1.1) indeed defines a positive (p,p)-current h ddcu i j j on X. We will show below that it is automatically closed (Theorem 1.8). V Condition (1.2) is always satisfied when p = 1, and in fact it is not difficult to show that hddcui= 1 ddcu. {u>−∞} There are however examples where non-pluripolar products are not well-defined as soon as p ≥ 2. This is most easily understood in the following situation. 6 S.BOUCKSOM,P.EYSSIDIEUX,V.GUEDJ,A.ZERIAHI Definition 1.2. A psh function u on X will be said to have small unbounded locus if there exists a (locally) complete pluripolar closed subset A of X outside which u is locally bounded. Assume that u ,...,u have small unbounded locus, and let A be closed com- 1 p plete pluripolar such that each u is locally bounded outside A (recall that com- j plete pluripolar subsets are stable under finite unions). Then h ddcu i is well- j j defined iff the Bedford-Taylor product ddcu , which is defined on the open j j V subsetX−A,haslocally finitemassneareach pointofA. Inthatcaseh ddcu i V j j is nothing but the trivial extension of ddcu to X. j j V Example 1.3. Consider Kiselman’s exVample (see [32]) u(x,y) := (1−|x|2)(−log|y|)1/2 for (x,y) ∈ C2 near 0. The function uispsh near 0, itis smooth outside the y = 0 axis, but the smooth measure (ddcu)2, defined outside y = 0, is not locally finite near any point of y = 0. This means that the positive Borel measure h(ddcu)2i is not locally finite. We now collect some basic properties of non-pluripolar products. Proposition 1.4. Let u ,...,u be psh functions on X. 1 p • The operator (u ,...,u )7→ h ddcu i 1 p j j ^ is local in the plurifine topology whenever well-defined. • The currenth ddcu iand thefactthatitiswell-definedboth onlydepend j j on the currents ddcu , not on the specific potentials u . j j V • Non-pluripolar products, which are obviously symmetric, are also multi- linear in the following sense: if v is a further psh function, then 1 h(ddcu +ddcv )∧ ddcu i= hddcu ∧ ddcu i+hddcv ∧ ddcu i 1 1 j 1 j 1 j j≥2 j≥2 j≥2 ^ ^ ^ in the sense that the left-hand side is well-defined iff both terms in the right-hand side are, and equality holds in that case. Proof. Let us prove the first point. If u and v are psh functions on X such j j that u = v on a given plurifine open subset O, then the locally bounded psh j j functions max(u ,−k) and max(v ,−k) also coincide on O. If we set j j E := {u > −k}∩ {v > −k}, k j j j j \ \ then we infer 1 ddcmax(u ,−k) = 1 ddcmax(v ,−k), O∩Ek j O∩Ek j j j ^ ^ hence in the limit 1 h ddcu i = 1 h ddcv i O j O j j j ^ ^ MONGE-AMPE`RE EQUATIONS IN BIG COHOMOLOGY CLASSES 7 as desired by Lemma 1.5 below. We now prove thesecond point. Letw bepluriharmoniconX, andlet K bea j compact subset. We can find C > 0 such that w ≤ C on an open neighborhood j V of K. On the plurifine open subset O := {u +w > −k}∩V ⊂ {u > −k−C}∩V k j j j j j \ \ the following locally bounded psh functions coincide: max(u +w ,−k) = max(u ,−w −k)+w = max(u ,−k−C)+w . j j j j j j j Since ddcw = 0, it follows that j ωn−p∧ ddcmax(u +w ,−k) = ωn−p∧ ddcmax(u ,−k−C) j j j ZOk j ZOk j ^ ^ ≤ ωn−p∧ ddcmax(u ,−k−C) j ZTj{uj>−k−C}∩V j ^ which is uniformly bounded in k by assumption, and the second point is proved. The proof of the last point is similarly easy but tedious and will be left to the reader. (cid:3) Lemma 1.5. Assume that the non-pluripolar product h ddcu i is well-defined. j j Then for every sequence of Borel subsets E such that k V E ⊂ {u > −k} k i j \ and X − E is pluripolar, we have k k S lim 1 ddcmax(u ,−k)i = h ddcu i k→∞ Ek j j j j ^ ^ against all bounded measurable functions. Proof. This follows by dominated convergence from (1 −1 ) ddcmax(u ,−k)i ≤ (1−1 )h ddcu i. ∩j{uj>−k} Ek j Ek j j j ^ ^ (cid:3) A crucial point for what follows is that non-pluripolar products of globally defined currents are always well-defined on compact Ka¨hler manifolds: Proposition 1.6. Let T ,...,T be closed positive (1,1)-currents on a compact 1 p K¨ahler manifold X. Then their non-pluripolar product hT ∧ ... ∧ T i is well- 1 p defined. Proof. Let ω be a Ka¨hler form on X. In view of the third point of Proposi- tion 1.4, upon adding a large multiple of ω to the T ’s we may assume that their j cohomology classes are Ka¨hler classes. We can thus find Ka¨hler forms ω and j ω -psh functions ϕ such that T = ω +ddcϕ . Let U be a small open subset of j j j j j 8 S.BOUCKSOM,P.EYSSIDIEUX,V.GUEDJ,A.ZERIAHI X on which ω = ddcψ , where ψ ≤ 0 is a smooth psh function on U, so that j j j T = ddcu on U with u := ψ +ϕ . The bounded psh functions on U j j j j j ψ +max(ϕ ,−k) j j and max(u ,−k) j coincide on the plurifine open subset {u >−k} ⊂ {ϕ > −k}, thus we have j j ωn−p∧ ddcmax(u ,−k) j ZTj{uj>−k} j ^ = ωn−p∧ (ω +ddcmax(ϕ ,−k)) j j ZTj{uj>−k} j ^ ≤ ωn−p∧ (ω +ddcmax(ϕ ,−k)). j j ZX j ^ But the latter integral is computed in cohomology, hence independent of k, and this shows that (1.2) is satisfied on U, qed. (cid:3) Remark 1.7. The same property extends to the case where X is a compact com- plex manifold in the Fujiki class, that is bimeromorphic to a compact Ka¨hler manifold. Indeed there exists in that case a modification µ : X′ → X with X′ compact Ka¨hler. Since µ is an isomorphism outside closed analytic (hence pluripolar) subsets, it is easy to deduce that hT ∧...∧T i is well-defined on X 1 p from the fact that hµ∗T ∧...∧µ∗T i is well-defined on X′, and in fact 1 p hT ∧...∧T i = µ hµ∗T ∧...∧µ∗T i. 1 p ∗ 1 p On the other hand it seems to be unknown whether finiteness of non-pluripolar products holds on arbitrary compact complex manifolds. Building on the proof of the Skoda-El Mir extension theorem, we will now prove the less trivial closedness property of non-pluripolar products. Theorem 1.8. Let T ,...,T be closed positive (1,1)-currents on a complex man- 1 p ifold X whose non-pluripolar product is well-defined. Then the positive (p,p)- current hT ∧...∧T i is closed. 1 p Proof. The result is of course local, and we can assume that X is a small neigh- borhood of 0 ∈ Cn . The proof will follow rather closely Sibony’s exposition of the Skoda-El Mir theorem [39], cf. also [18, pp. 159-161]. Let u ≤ 0 be a local potential of T near 0 ∈ Cn, and for each k consider the j j closed positive current of bidimension (1,1) Θ := ρ∧ ddcmax(u ,−k) k j j ^ and the plurifine open subset O := {u > −k} k j j \ MONGE-AMPE`RE EQUATIONS IN BIG COHOMOLOGY CLASSES 9 so that 1 Θ converges towards Ok k ρ∧hT ∧...∧T i 1 p by (1.1). Here ρ is a positive (n−p−1,n−p−1)-formwith constant coefficients, so that Θ has bidimension (1,1). It is enough to show that k lim d(1 Θ ) = 0 k→∞ Ok k for any choice of such a form ρ. Let also u := u , so that u ≤ −k outside O , and set j j k P w := χ(eu/k), k where χ(t) is a smooth convex and non-decreasing function of t ∈ R such that χ(t) = 0 for t ≤ 1/2 and χ(1) = 1. We thus see that 0 ≤ w ≤ 1 is a non- k decreasing sequence of bounded psh functions defined near 0 ∈ Cn with w = 0 k outside O and w → 1 pointwise outside the pluripolar set {u = −∞}. Finally k k let 0 ≤ θ(t)≤ 1 be a smooth non-decreasing function of t ∈ R such that θ(t)= 0 for t ≤ 1/2 and θ ≡ 1 near t = 1. The functions θ(w ) are bounded, non- k decreasing in k and we have θ(w ) ≤1 k Ok since θ(w ) ≤ 1 vanishes outside O . Note also that θ′(w ) vanishes outside O , k k k k and converges to 0 pointwise outside {u = −∞}. Our goal is to show that lim d(1 Θ ) = 0. k→∞ Ok k But we have 0≤ (1 −θ(w ))Θ ≤(1−θ(w ))hT ∧...∧T i, Ok k k k 1 p andthelatterconvergesto0bydominatedconvergencesinceθ(w ) → 1pointwise k outsidethepolarsetofu,whichisnegligibleforhT ∧...∧T i. Itisthusequivalent 1 p to show that lim d(θ(w )Θ ) = 0. k k k→∞ Since w is a bounded psh function, Lemma 1.9 below shows that the chain rule k applies, that is d(θ(w )Θ ) = θ′(w )dw ∧Θ . k k k k k Recall that dw ∧Θ has order 0 by Bedford-Taylor, so that the right-hand side k k makes sense. Now let ψ be a given smooth 1-form compactly supported near 0 and let τ ≥ 0 be a smooth cut-off function with τ ≡ 1 on the support of ψ. The Cauchy-Schwarz inequality implies 2 θ′(w )ψ∧dw ∧Θ ≤ τdw ∧dcw ∧Θ θ′(w )2ψ∧ψ∧Θ . k k k k k k k k (cid:12)Z (cid:12) (cid:18)Z (cid:19)(cid:18)Z (cid:19) (cid:12) (cid:12) B(cid:12)ut on the one hand we(cid:12)have (cid:12) (cid:12) 2 τdw ∧dcw ∧Θ ≤ τddcw2 ∧Θ k k k k k Z Z 10 S.BOUCKSOM,P.EYSSIDIEUX,V.GUEDJ,A.ZERIAHI = w2ddcτ ∧Θ = w2ddcτ ∧Θ k k k k Z ZOk since w vanishes outside O , and the last integral is uniformly bounded since k k 0≤ w ≤ 1 and 1 Θ has uniformly boundedmass by (1.2). On the other hand k Ok k we have θ′(w )2Θ = θ′(w )21 Θ ≤ θ′(w )2hT ∧...∧T i k k k Ok k k 1 p since θ′(w ) also vanishes outside O , and we conclude that k k lim θ′(w )2ψ∧ψ∧Θ = 0 k k k→∞ Z by dominated convergence since θ′(w ) → 0 pointwise ouside the polar set of u, k which is negligible for hT ∧...∧T i. The proof is thus complete. (cid:3) 1 p Lemma 1.9. Let Θ be a closed positive (p,p)-current on a complex manifold X, f be a smooth function on R and v be a bounded psh function. Then we have d(f(v)Θ) = f′(v)dv∧Θ. Proof. Thisisalocalresult,andwecanthusassumethatΘhasbidimension(1,1) by multiplying it by constant forms as above. The result is of course true when v is smooth. As we shall see the result holds true in our case basically because v belongs to the Sobolev space L2(Θ), in the sense that dv∧dcv∧Θ is well-defined. 1 Indeed the result is standard when Θ = [X], and proceeds by approximation. Here we let v be a decreasing sequence of smooth psh functions converging k pointwise to v. We then have f(v )θ → f(v)Θ by dominated convergence, thus k it suffices to show that lim f′(v )dv ∧Θ = f′(v)dv∧Θ. k k k→∞ We write f′(v )dv ∧Θ−f′(v)dv∧Θ = f′(v )−f′(v) dv ∧Θ+f′(v)(dv −dv)∧Θ. k k k k k Let ψ be a test 1-form and τ ≥ 0 be a smooth cut-off function with τ ≡ 1 on the (cid:0) (cid:1) support of ψ. Cauchy-Schwarz implies 2 f′(v )−f′(v) ψ∧dv ∧Θ k k (cid:12)Z (cid:12) (cid:12) (cid:0) (cid:1) (cid:12) ≤ f′(vk(cid:12)(cid:12))−f′(v) 2ψ∧ψ∧Θ τdvk(cid:12)(cid:12)∧dcvk ∧Θ . (cid:18)Z (cid:19)(cid:18)Z (cid:19) The second factor(cid:0)is bounded sin(cid:1)ce dv ∧dcv ∧Θ converges to dv∧dcv∧Θ by k k Bedford-Taylor’s monotone convergence theorem, and the first one converges to 0 by dominated convergence. We similarly have 2 f′(v)(dv −dv)∧ψ∧Θ k (cid:12)Z (cid:12) (cid:12) (cid:12) ≤ f′(v)2ψ∧(cid:12)(cid:12) ψ∧Θ τd(vk −v)∧d(cid:12)(cid:12)c(vk −v)∧Θ , (cid:18)Z (cid:19)(cid:18)Z (cid:19) wherenowthefirstfactor is boundedwhilethesecondonetends to0by Bedford- Taylor once again, and the proof is complete. (cid:3)
Description: